Using Visual Design Expertise to Characterize the Effectiveness of 2D Scientific Visualization Methods

Figure 1: Eleven different visualization methods that represent the same continuous scalar dataset. We are characterizing the effectiveness of each one of these methods, both individually and in combination, to represent scalar datasets in 2D. We present the results from a pilot study that evaluates the effectiveness of 2D visualization methods in terms of a set of design factors, which are subjectively rated by expert visual designers. In collaboration with educators from the Illustration Department at the Rhode Island School of Design (RISD), we have defined a space of visualization methods using basic visual elements including icon hue, icon size, icon density, and background saturation (see Figure 1). In this initial pilot study we presented our subjects with single variable visualization methods. The results characterize the effectiveness of individual visual elements according to our design factors. We are beginning to test these results by creating two-variable visualizations and studying how the different visual elements interact. 1 INTRODUCTION Given the increasing capacity of scientists to acquire or calculate multival-ued datasets, creating effective visualizations for understanding and correlating these data is imperative. However, modeling the space of possible vi-sualization methods for a given scientific problem has challenged computer scientists, statisticians, and cognitive scientists for many years [1,2,3,4]; it is still an open challenge. Our goal is to provide scientists with visualization methods that convey information by optimizing the design of the images to facilitate perception and comprehension. We created a framework for evaluating these visualization methods through feedback from expert visual designers and art educators. Our framework mimics the art education process, in which art educators impart artistic and visual design knowledge to their students through critiques of the students' work.We established a set of factors that characterize the effectiveness of a visualization method in displaying scientific data. These factors include constraints implied by the dataset, such as the relative importance of the different data variables or the minimum feature size present in the data. We also include design, artistic, and perceptual factors, such as time required to understand the visualization, or how visually linear is the mapping between data and visual element across the image. We will describe these in detail in section 2. Evaluating the effectiveness of visualizations is difficult because tests to evaluate them meaningfully are hard to design and execute [5]. We have researched this issue previously in two user studies comparing 2D vector visualization methods. The first …